INDEX

Explanations

terms related to weakening, discrediting, or harming something or someone

oai_token-act-pair · gpt-3.5-turbo

instances of the word "undermine" and related concepts of subversion or damage to integrity

oai_token-act-pair · gpt-4o-mini Triggered by @bot

New Auto-Interp

Top Features by Cosine Similarity

Comparing With GPT2-SMALL @ 5-res-jb

Configuration

jbloom/GPT2-Small-SAEs-Reformatted/blocks.5.hook_resid_pre

Prompts (Dashboard)

24,576 prompts, 128 tokens each

Dataset (Dashboard)

Skylion007/openwebtext

Features

24,576

Data Type

torch.float32

Hook Point

blocks.5.hook_resid_pre

Architecture

standard

Context Size

128

Dataset

Skylion007/openwebtext

Hook Point Layer

Activation Function

relu

Embeds

PlotsExplanationShow Test FieldDefault Test Text

IFrame

Link

Not in Any Lists

No Comments

Negative Logits

ëĭ

-0.67

cas

-0.61

lov

-0.61

enne

-0.61

////////

-0.61

ann

-0.59

andro

-0.59

cise

-0.59

Flo

-0.59

 Reflex

-0.58

POSITIVE LOGITS

 havoc

0.84

 undermin

0.83

ments

0.75

ingly

0.72

lessly

0.71

 undermining

0.69

 erosion

0.68

 guiActiveUn

0.68

amental

0.65

ionage

0.65

Activations Density 0.041%

terms related to weakening, discrediting, or harming something or someone

instances of the word "undermine" and related concepts of subversion or damage to integrity

No Comments

No Known Activations